The California Health Interview Survey (CHIS) is a population-based random digit dial telephone survey conducted biennially since 2001 [11]. The CHIS is designed to provide cross-sectional, population-based statewide estimates of health indicators for all major racial and ethnic groups as well as several Asian ethnic subgroups. The CHIS obtains oral informed consent from each respondent prior to conducting the survey. This analysis was exempt from Institutional Review Board (IRB) review by the University of California, Los Angeles IRB because it was based on existing data and information was recorded in such a manner that subjects could not be identified. Data analysis was conducted in 2009. We merged data from the 2001, 2003, and 2005 surveys and created an analysis dataset consisting of adults 50 years of age and older with no history of CRC who self-identified as Chinese (N = 1,432), Filipino (N = 753), Vietnamese (N = 709), Korean (N = 675) or Japanese American (N = 619). Since Asian Americans make up 12.5% of the population in California, which is a much larger proportion than the national average of 4.5% (http://quickfacts.census.gov/qfd/states/06000.html, accessed 12/15/2009), and since CHIS oversampled several Asian American groups, this data set is well suited to examining disparities in CRC screening among Asian American ethnic groups.
Outcome variable
Our primary outcome variable was "ever received CRC screening," a binary variable. Respondents with an affirmative response to the question "Have you ever done a blood stool test, using a home test kit?" and/or an affirmative response to the question "Have you ever had a sigmoidoscopy or colonoscopy?" (in 2001 and 2003, this question also included proctoscopy) were classified as ever received CRC screening. Respondents who responded negatively to both questions were classified as never received CRC screening. Since the reason for screening was not assessed in all three years, we could not restrict the analysis to routine screening; hence results pertain to CRC screening for any reason. We used "ever received CRC screening" rather than "screened according to the guidelines" as the outcome because type of exam and time frame were not assessed in the same manner in all three years. In addition, using "ever received CRC screening" allowed us to obtain larger numbers than "screened according to the guidelines," which had small numbers in some racial/ethnic groups, and is less subject to faulty recall of timing of tests.
Statistical analysis
Prior to conducting multivariate analyses, we examined the bivariate relationship of past CRC screening with each of our potential explanatory variables. For each Asian American group, we estimated the proportion with past CRC screening by year (2001, 2003, 2005), age (50-64 yr, ≥ 65 yr), gender, education (≤ high school, > high school), marital status (yes, no), employment status (yes, no), percent federal poverty level (0-99%, 100-199%, 200-299%, ≥ 300%), English proficiency, health insurance (yes, no), usual source of care (yes, no) and most recent doctor's visit (≤ 1 year, > 1 year). In the CHIS, English proficiency ("How well do you speak English?") was asked of non-U.S. born respondents only; we dichotomized the responses as inapplicable because U.S. born/very well versus well/not well/not at all well. Proportions were estimated using the proportion procedure for survey data (svy proportion) in Stata Version 9.1 with CHIS-provided survey sampling weights, so that the estimates are applicable to the California population of the five Asian ethnic groups.
Multivariate models for the outcome variable "ever received CRC screening" were fit using the logistic regression procedure for survey data (svy logistic) in Stata Version 9.1 with CHIS-provided survey sampling weights. Guided by the Andersen Behavioral Model of Health Services Use [12], which suggests that people's use of health services is determined by predisposing characteristics (e.g., demographics), enabling resources (e.g., language proficiency and access to health care) and health beliefs, we added the variables in this stepped approach to understand the impact of predisposing characteristics (demographics) and enabling resources (e.g., language proficiency and access to health care) on CRC screening and to test if screening disparities are fully explained by these factors. We fit four models, a base model (Model 1) estimating trends in screening prevalence among the five Asian ethnic groups adjusted for age and gender, and three models that added covariates in blocks, sequentially adjusting for demographic characteristics (Model 2), English proficiency (Model 3), and access to health care (Model 4), with each model retaining the variables in the previous model. Model 1 included as covariates Asian ethnic subgroup (Chinese, Filipino, Vietnamese, Korean, Japanese), year of survey (2001, 2003, 2005), interactions between subgroup and year, age and gender. Model 2 added education, marital status, employment status, and percent federal poverty level. Model 3 added English proficiency. Model 4 added access to health care as measured by health insurance, usual source of care and most recent doctor's visit. The joint significance of each block of added covariates was assessed using adjusted Wald tests. Health beliefs were not assessed by the CHIS in a manner that would permit inclusion in the multivariate models. However, we did perform a separate analysis of cited reasons for non-adherence to CRC screening (see below).
Japanese Americans, who had the highest screening prevalence, and the year 2001 were used as reference categories for ethnic subgroup and year, respectively. For the main effect of subgroup, the odds ratios compare each subgroup to Japanese Americans in 2001. For the main effect of year, the odds ratios compare 2003 to 2001 and 2005 to 2001 for Japanese Americans. The subgroup-by-year interaction terms allow for trends that differ from Japanese Americans in the other subgroups. The odds ratios for the interactions of subgroup with 2003 (or 2005) are ratios of odds ratios that compare the trend in each subgroup from 2001 to 2003 (or 2005) to the trend in Japanese Americans from 2001 to 2003 (or 2005). For these interaction terms, an odds ratio less than unity indicates a negative trend, while an odds ratio greater than unity indicates a positive trend compared to Japanese Americans.
We obtained a nonparametric estimate of the relationship between the log odds of ever receiving screening and age using a lowess curve, in which each smoothed value is obtained as a weighted quadratic least squares regression over a span of values of the y-axis variable, using the Stata lowess command. This curve showed that the log odds of ever receiving CRC screening increased with age until about 73 years of age and then declined (results not shown). To model this relationship, we included a quadratic term for age in the models and centered this variable.
Reasons for not receiving CRC screening
In 2001 and 2005, respondents whose most recent endoscopy was more than 10 years ago or who had never had one were asked "What is the one most important reason why you (never had/not had) one of these exams (in the past 10 years)?" Respondents whose most recent FOBT was more than one year ago or who had never had one were asked "What is the most important reason you have (never had/not had) a home blood stool test (in the past 12 months)?" We grouped reasons for not receiving CRC screening into four categories: unaware of test, have no health problems, fear of pain/embarrassed (endoscopy only) and other reasons, then cross-classified the responses by Asian subgroup. We tested for differences in the distribution of reasons by ethnicity using chi-square tests of the null hypothesis of homogeneity, and identified cells with unusually high or low numbers under the null hypothesis using the criterion of a standardized residual with absolute value greater than 3 [13]. Data from 2001 and 2005 were combined for this analysis.
Doctor recommendation
In 2001 and 2005, respondents whose most recent endoscopy was more than 10 years ago or who had never had one were asked "During the past 12 months, has a doctor recommended that you have a sigmoidoscopy or colonoscopy?" Respondents whose most recent FOBT was more than one year ago or who had never had one were asked "In the past 12 months, has a doctor recommended that you have a home blood stool test?" Based on these questions, we created a binary variable "Doctor recommended endoscopy and/or FOBT in the past 12 months." For each of the five Asian subgroups, we computed the proportion who had received a doctor's recommendation in 2001 and 2005, respectively. We tested for differences among the Asian subgroups within each year and tested for change over time within each subgroup using two-sample tests for differences of proportions.
The CHIS questions regarding reasons for not receiving screening and doctor recommendation were only applicable to the subset of respondents who were non-adherent to CRC screening, and the number of relevant respondents in some subgroups (e.g., Japanese Americans) was relatively small. For these reasons, we conducted unweighted analyses of these data, since proper use of the survey weights was not feasible.